Coal cutter navigation cutting planning method and system based on working face step model

Abstract

The invention discloses a coal cutter navigation cutting planning method and system based on a working face step model, and the method comprises the steps: directly representing the step mutation of a coal seam floor through constructing a discretized working face step model; dividing a three-level geological cognitive state based on the completed cutter number and the predicted sight distance, realizing spatial quantization of geological information uncertainty, and establishing a credible geological model dynamic evolution mechanism from prior driving to data driving; on this basis, the floor height is modeled as a first-order Markov hidden state variable, probability distribution of a future floor profile is generated by adopting Kalman filtering and multi-step Bayesian prediction, and online compensation is carried out by combining exponential decay weighting historical deviation; and finally, by taking the prediction profile as a main track and the height adjustment compensation amount output by the cutting current layering system as feedback correction, generating the target height of the roller through adaptive weighted fusion, and realizing collaborative optimization of geological cognition, multi-step prediction and autonomous height adjustment.

Description

Technical Field

[0001] This invention belongs to the field of coal mining technology, and specifically relates to a method and system for planning and cutting of coal mining machines based on a working face step model. Background Technology

[0002] Intelligent coal mining is a core direction for the industry's development. Unmanned operation of fully mechanized mining faces requires coal mining machines to possess autonomous perception and adaptive cutting capabilities based on coal seam geological conditions. Achieving autonomous and stable cutting under complex geological conditions is the core objective of intelligent mining. Early coal mining machine cutting mainly relied on memorized cutting patterns, recording a standard cutting trajectory through manual teaching, which was then repeated in subsequent cuts. This method is applicable to stable coal seam conditions but struggles to adapt to variations in coal seam undulations. Under complex coal seam conditions, it often leads to undercutting or overcutting, resulting in resource waste or equipment damage.

[0003] Existing adaptive cutting trajectory planning methods for coal mining machines still have the following shortcomings in practical applications: (1) Most existing methods rely on pre-set geological models or historical data for trajectory planning. However, under complex coal seam conditions, geological conditions are significantly concealed and change suddenly. Due to the harsh underground environment, the real-time performance and accuracy of various sensing methods are difficult to guarantee, making it difficult for the system to respond to unforeseen geological changes in a timely manner.

[0004] (2) Most existing methods focus on optimizing the geological model and trajectory planning, and pay insufficient attention to the physical response of the cutting process. When the coal mining machine encounters interbedded rock, hard coal or sudden changes in rock strata during actual cutting, it lacks an autonomous height adjustment decision mechanism based on real-time physical performance data such as cutting current and vibration. The drum height adjustment mostly relies on preset trajectories or offline strategies, which cannot respond to sudden load changes in time, resulting in the risk of over-cutting, under-cutting or equipment overload, which restricts the autonomous adaptability of the cutting process.

[0005] (3) Most methods adopt the model of "planning the trajectory of the next cut based on the data of the previous cut", which is essentially a post-correction rather than real-time adaptation. When the coal seam changes drastically in strike or dip, this kind of lagging planning cannot achieve true on-demand adjustment, and there is still a risk of undercutting or overcutting in the cutting process.

[0006] (4) Existing technologies attempt to integrate multi-dimensional data such as virtual pre-planning, real-time perception, and device pose, but lack a dynamic trade-off mechanism based on confidence level when information conflicts occur. The system has difficulty in determining whether to rely on the macroscopic model or the microscopic detection, resulting in uncertainty in control decisions.

[0007] (5) Existing methods are usually modeled or trained for specific working faces, and the models have strong scene dependence. When the coal mining machine is transferred to coal seams with different geological conditions, the adaptability of the original model decreases significantly, and data needs to be collected again for calibration or training. It lacks the ability to transfer knowledge across working faces. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for coal mining machine navigation and cutting planning based on a working face step model, which at least solves the technical problems of difficulty in effectively representing the step changes in the coal seam floor under complex coal seam conditions and the lack of dynamic adaptability of the planned path.

[0009] To achieve the above objectives, according to one aspect of the present invention, a method for coal mining machine navigation and cutting planning based on a working face step model is provided, comprising: Step 1: Construct a working face stepped model and a three-level geological cognitive state model, including: The coal seam floor is discretized along the advancing direction into a sequence of transverse profiles corresponding to each cut. Each transverse profile consists of a set of discrete bottom points distributed along the cutting direction, forming a stepped model of the working face. Based on the current number of cuts completed and the preset predicted sight distance, a three-level geological cognition state model is constructed, dividing the working face into three cognition zones along the advancement direction: The high-confidence measurement area is the area that has been actually cut and for which data collection has been completed. The Zhongzhixin extrapolation area is the range that has not yet been cut off but can be predicted by extrapolation from historical profiles. The low-confidence blind probe zone is a distant region that exceeds the current system's cognitive capabilities; As the coal mining machine advances, the boundaries of the three cognitive regions move forward synchronously, realizing the dynamic migration of cognitive states; Step two, establish the dynamic evolution mechanism of a credible geological model, including: Based on multi-source static geological data, an initial discretized three-dimensional base plate height field is constructed as a priori model. After each cut is completed, the measured data is interpolated and mapped to the grid cells of the prior model. For grid cells covered by the measured data, a parameter coverage operation is performed and the confidence state is updated. Step 3 involves performing multi-step probability prediction and real-time updates based on Bayesian state estimation, including: Based on historical actual trajectories and prior geological models, the bottom plate height is modeled as a first-order Markov hidden state variable. The posterior distribution is solved recursively using Kalman filtering, and the optimal estimate and variance of the current base plate height are output. Multi-step Bayesian prediction based on state transition model is used to generate the probability distribution of future H-blade bottom plate height, and boundary correction is performed using the measured height of the roadway. An outer loop real-time update mechanism with exponential decay weighted historical bias is introduced to compensate for the prediction online and output a future base prediction profile with probability information. Step four involves performing dynamic cutting of the base plate and autonomous height adjustment control, including: By integrating multi-sensor data to calculate the roller pose in real time, incremental modeling is used to generate a six-degree-of-freedom dynamic cutting block, forming a continuously evolving dynamic base plate model. Using the Bayesian prediction profile as the main trajectory and the height adjustment compensation amount output based on the cutting current hierarchical system as feedback correction, the target height of the roller is generated through adaptive weighted fusion, and the height adjustment command is issued after kinematic constraints and timing verification.

[0010] As a preferred implementation, the method includes performing sensor error simulation and system verification, constructing an error-prone coal mining machine model, superimposing a preset error including inertial navigation drift, random noise and tilt sensor error onto the theoretical real pose, generating a simulated sensor data-driven digital twin scene; and evaluating the ability to suppress sensor noise by comparing the trajectory deviation and height adjustment response of the theoretical coal mining machine and the error-prone coal mining machine.

[0011] In a preferred implementation, in step one, the working face step model is defined as an ordered set: ; Where, x k Let s be the advancing position corresponding to the k-th cut. k The cross section corresponding to the k-th cut can be represented as defined in [z s ,z e Functions on ]; The working face stepped model uses actual cutting actions as an index, with no continuity constraints between adjacent sections, directly reflecting local mutations and providing a structural basis for subsequent cognitive state division.

[0012] In a preferred embodiment, in step two, the parameter coverage operation replaces the original prior height value of the grid cell with the measured height value, and at the same time replaces the original prior variance with the sensor measurement variance. For the blind grid that is not covered, the original prior geological data is retained.

[0013] In a preferred implementation, in step two, updating the confidence status involves transitioning the region that has been truncated and filled with measured data from the medium-confidence projection region to the high-confidence measured region, and marking it as "credible"; the newly exposed remote region is moved from the low-confidence blind probe region to the medium-confidence projection region and marked as "to be verified".

[0014] In a preferred implementation, step three, the multi-step probability prediction based on Bayesian state estimation, specifically involves: At a fixed cutting position z, the bottom plate height is modeled as a hidden state variable that evolves with the number of cuts k. x k ( z Considering the random fluctuations in the bottom plate height between adjacent cuts, a first-order Markov process is used to describe its state evolution, and the state evolution equation is established: ; Among them, among them, w k ( z () represents process noise, used to characterize the random variation in bottom plate height between adjacent cuts caused by local nonstationarity of geological structures; Establish observation equations based on state evolution equations: ; Among them, A k (z) represents the observed actual cutting trajectory of the k-th cut. v k ( z () represents observation noise, used to characterize sensor measurement error; The posterior distribution of the state is solved recursively using Kalman filtering to obtain the posterior mean. and posterior variance ; Based on the state transition model, the probability prediction of the height of the (k+m)th blade base plate is as follows: ; Their mean constitutes the preliminary prediction profile. ; ; Introducing an uncertainty threshold and high mutation threshold If the posterior variance Or adjacent knife difference amplitude value If the predicted gradient at position z is zero, then the predicted gradient at position z will be set to zero.

[0015] In a preferred implementation, step three, the outer ring real-time update mechanism specifically includes: After each cut is completed, calculate the actual bottom plate trajectory A of that cut. k (z) and the final predicted profile P used before cutting. k Deviation between (z): The deviation sequence is stored in the historical buffer. ; When generating predictions for the nth subsequent cut, an exponentially decaying weight is used to weight and fuse historical biases to obtain a compensation term. c n (z ); ; Among them, weight λ is the attenuation coefficient, and the compensation term is added to the basic prediction P. n On (z), the final predicted trajectory is obtained after real-time updates.

[0016] In a preferred embodiment, step four, the current stratification system includes: Establish a multi-level threshold mapping table between the cutting current range and the coal and rock firmness coefficient, subdivide the current range into multiple levels, and different levels correspond to different adjustment strategies; A time-duration criterion is introduced to determine the operating condition based on the current value and its duration. The current threshold range is corrected by recording the actual cutting effect after the current alarm.

[0017] In a preferred embodiment, in step four, adaptive weighted fusion is used to generate the target height h of the roller. target ( z ); ; Among them, P k+1 (z) is the predicted profile, and Δh is the adjustment compensation amount output by the current stratification system; a ( z ) is an adaptive weighting coefficient, which is determined by the current posterior variance of the base plate and the current value.

[0018] According to another aspect of the present invention, a coal mining machine navigation and cutting planning system based on a working face step model is provided, which performs the above-described method.

[0019] This invention constructs a stepped working face model, discretizing the continuous coal seam floor into a sequence of transverse profiles aligned with the cutter strokes. This discretized structure naturally represents the step-like abrupt changes in the coal seam, avoiding the smooth masking of local abrupt changes by traditional continuous smooth models. Through a three-tiered geological cognitive state model—"high-confidence measured area, medium-confidence projected area, and low-confidence blind exploration area"—the uncertainty of geological information in different regions is quantified. A reliable geological model update mechanism is established that dynamically evolves with the progress of the coal mining machine, achieving a gradual cognitive evolution from prior-driven to data-driven approaches.

[0020] This invention generates a probability distribution of the future multi-blade base plate height based on Bayesian state estimation, quantifying the uncertainty of prediction; combined with an exponentially decaying weighted historical deviation real-time update mechanism, it achieves adaptive online correction of the prediction model, effectively suppressing long-term drift.

[0021] This invention maps the amplitude and fluctuation characteristics of the cut-off current to the adjustment compensation amount through a current stratification system, enabling the system to respond in real time to local geological disturbances such as hard rock and interbedded rock; the adaptive weighted fusion mechanism dynamically balances the weights of the predicted trajectory and the current feedback.

[0022] This invention ensures that the height adjustment command is physically achievable by applying kinematic constraints to the target height and verifying the feasibility of the height adjustment timing, thus avoiding control failures caused by limitations in the capabilities of the actuator.

[0023] This invention forms a complete closed-loop architecture, constructing a complete intelligent cutting closed loop of "geological cognition - multi-step prediction - dynamic execution - feedback correction". Relying on the digital twin platform, it realizes real-time synchronous mapping between the physical system and the virtual model, effectively improving the adaptive cutting capability of the coal mining machine under complex geological conditions and the response speed to geological changes. Attached Figure Description

[0024] Figure 1 This is a block diagram of the overall construction of the coal mining machine navigation and cutting planning method based on the working face step model described in this invention; Figure 2 This is a ladder model and a cognitive evolution diagram; Figure 3 Block diagram of Bayesian multi-step prediction and real-time update module; Figure 4 A control block diagram for dynamic base plate generation and height adjustment; Figure 5 This is a block diagram for sensor error simulation and system verification. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the present invention will be further described clearly and completely below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.

[0026] The basic concept of this invention is to use the working face stepped model as the base of discretized data, divide the coal seam information into spatial confidence levels through a three-level geological cognitive state model, and establish a dynamic evolution mechanism of the credible geological model to achieve the gradual integration of prior knowledge and measured data. On this basis, Bayesian state estimation is used to perform multi-step probability prediction and real-time updates. Finally, physical perception information such as cutting current is integrated to perform autonomous height adjustment control, and a dynamic coupling mechanism between the cutting trajectory and the geological model is established to achieve the synergistic optimization of geological cognition, multi-step prediction and autonomous height adjustment.

[0027] The protection technical solution claimed by the present invention will be further clearly and completely described below with reference to a relatively specific embodiment.

[0028] 1. Construct a working face stepped model and a three-level geological cognitive state model. This step constructs the data foundation for the navigation and cutting planning. A three-level coordinate system and a spatial benchmark are established; the coal seam floor is discretized into a sequence of transverse profiles aligned with the cuts, forming a working face step model represented by discrete point sets, which can directly describe the abrupt changes in the coal seam. Based on this, a three-level geological cognitive state is defined based on the completed cuts and predicted sight distances: "high-confidence measured area, medium-confidence projected area, and low-confidence blind exploration area," realizing the spatial quantification of geological information uncertainty.

[0029] In a relatively specific embodiment, this step includes steps 101-104.

[0030] Step 101: Establish a three-level coordinate system for the fully mechanized mining face, including the global navigation coordinate system OX. b Y b Z b Working surface coordinate system OX n Y n Z n and the coal mining machine carrier coordinate system OX s Y s Z s .

[0031] The global navigation coordinate system is fixed at the reference point of the underground roadway and is used to determine the absolute position of the working face; the local coordinate system of the working face is established along the strike and dip of the working face and is used to describe the geological characteristics of the coal seam; the coordinate system of the coal mining machine carrier is fixed to the body of the coal mining machine and is used to calculate the relative pose of the drum.

[0032] By establishing a homogeneous transformation matrix between three coordinate systems, the fusion of multi-source sensor data under a unified spatial reference is achieved, providing an accurate spatial reference for subsequent pose calculation and trajectory planning, and eliminating the cumulative error caused by the inconsistency of coordinate systems.

[0033] Step 102: Discretize the coal seam floor along the advancing direction into a series of transverse profiles that correspond one-to-one with the cutting cutter of the coal mining machine. Each transverse profile consists of a set of discrete bottom points distributed along the cutting direction, forming a working face step model S.

[0034] Let the initial position of the coal mining machine be ( x 0, y 0, z 0), with a cut-off depth of d, then the advancing position corresponding to the k-th cut is x. k =x0+(k-1)d; At this position, the height distribution of the bottom plate along the cutting direction forms a transverse profile, and the effective range of the fully mechanized mining face in the cutting direction is [z s ,z e ], where z s and z eCorresponding to the boundary positions on the side of the haulage roadway and the return airway respectively; the cross-section corresponding to the k-th cut can be expressed as a function s s ,z e defined on [z k , and stored in the form of a discrete point set . On this basis, the working face step model is formally defined as an ordered set: ; This model is indexed by the actual cutting action. There is no continuity constraint between adjacent cross-sections, and local mutations can be directly reflected in the model, providing a clear structural basis for subsequent cognitive state division.

[0035] Step 103: Based on the current completed number of cuts k and the preset prediction sight distance H, divide the working face along the advancing direction into three cognitive regions: a high-confidence measured area, a medium-confidence deduced area, and a low-confidence blind exploration area.

[0036] The high-confidence measured area corresponds to the position where x < x0 + (k - 1)d, which is the area where actual cutting has been completed and data collection has been carried out; the medium-confidence deduced area covers the interval x0 + (k−1)d ≤ x < x0 + (k + H−l)d, which is the range that has not been cut but can be predicted by extrapolating historical cross-sections; the low-confidence blind exploration area corresponds to the area where x ≥ x0 + (k + H−l)d, which is the distal area beyond the current system's cognitive ability; where x0 is the initial advancing position of the coal cutter and d is the cutting depth.

[0037] Step 104: As the k-th cut is completed, update the current number of cuts k to k + 1, and the boundaries of the three cognitive regions move forward synchronously by one cutting depth unit; the part of the original medium-confidence deduced area close to the coal cutter is transferred to the high-confidence measured area due to the completion of cutting, and the position adjacent to the front edge of the medium-confidence deduced area in the original low-confidence blind exploration area is included in the medium-confidence deduced area, realizing the dynamic migration of the cognitive state.

[0038] 2. Establishing a dynamic evolution mechanism for a reliable geological model This module defines the dynamic migration rules of geological cognitive states with the advancement of the coal cutter. An initial three-dimensional floor height field is constructed based on multi-source static data as a prior model; every time a cut is completed, the measured data is interpolated and mapped to grid cells, and the covered area is replaced with the measured value and measurement variance. The completed cutting area jumps from the medium-confidence deduced area to the high-confidence measured area, and the newly exposed distal area is included in the low-confidence blind exploration area, realizing the gradual evolution of the geological model from "prior-driven" to "data-driven" with the advancement of the number of cuts.

[0039] In a relatively specific embodiment, this step includes the following steps 201 and 202.

[0040] Step 201: Based on multi-source static geological data, after data cleaning and interpolation, the working face is divided into regular grid-like coal seam sub-blocks along the cutting direction and the advancing direction; the center height and local attitude parameters of each sub-block are calculated using interpolation methods to construct an initial discretized three-dimensional floor height field.

[0041] The multi-source static geological data includes borehole data, tunnel exposure data, and seismic exploration data.

[0042] Step 202, the dynamic evolution mechanism of the geological model based on cognitive partitioning, specifically includes the following four sub-steps: Step 202-1: After the coal mining machine completes the k-th cut, the system maps the collected measured floor height data onto the grid nodes of the initial geological model. Since the measured points are discretely distributed while the model grid is regularly distributed, the inverse distance weighted interpolation (IDW) method is used to allocate the measured point data to the nearest grid cell, ensuring that the measured information is accurately filled into the corresponding position in the model.

[0043] Step 202-2: For grid cells covered by measured data, perform a parameter coverage operation. Replace the original prior height values ​​of the grid with the measured height values, and replace the original prior variance with the sensor measurement variance (usually much smaller than the prior variance). For uncovered blind grid areas, retain the original prior geological data to ensure model integrity.

[0044] Step 202-3 updates the confidence status of grid cells based on their data sources. Areas that have been truncated and populated with measured data transition from a medium-confidence projection area to a high-confidence measured area, and are marked as "credible." Newly exposed distant areas are moved from low-confidence blind exploration areas to medium-confidence projection areas and marked as "to be verified." This transition transforms the geological model from "empirical speculation" to "factual evidence."

[0045] In step 202-4, as the cut number k is updated to k+1, the spatial boundaries of the three confidence partitions are simultaneously advanced by one cutting depth. The system automatically releases the high-confidence region data that has moved far from the working face to save memory, while loading the prior data of the low-confidence region ahead into the prediction queue, realizing the "rolling update" and "infinite extension" of the geological model to adapt to the continuous advancement requirements of the longwall working face.

[0046] 3. Perform multi-step probability prediction and real-time updates based on Bayesian state estimation. This module is the core algorithm layer for navigation cut-off planning. The floor height is modeled as a first-order Markov hidden state variable, and the posterior distribution is solved recursively using Kalman filtering to output the optimal estimate and variance of the current floor height. Multi-step Bayesian prediction is performed based on the state transition model to generate the probability distribution of the future H-cutter floor height, and boundary corrections are made using the measured height of the tunnel. Simultaneously, an exponentially decaying weighted historical bias outer-loop real-time update mechanism is introduced to compensate for the prediction online, outputting a future floor prediction profile with probabilistic information.

[0047] In a relatively specific embodiment, this step includes steps 301-305.

[0048] Step 301: After completing the k-th cut, the system constructs a prediction of the future H-cut bottom plate profile based on the historical actual trajectory and prior geological model; at a fixed cut position z, the bottom plate height is modeled as a latent state variable x that evolves with cut k. k (z), considering the random fluctuations in the bottom plate height between adjacent cuts, a first-order Markov process is used to describe its state evolution law, and the state evolution equation is established: ; in, w k ( z The noise is process noise, used to characterize the random variation in floor height between adjacent cuts caused by local nonstationarity of geological structures. This noise follows a zero-mean Gaussian distribution, and its variance is... The variance of the sample can be estimated online using historical cut-off residuals or preset based on the geological report of the mining area.

[0049] The rationale for choosing a Markov process is that the coal seam floor typically exhibits continuity between adjacent cuts, meaning the height of the current cut depends primarily on the height of the previous cut, with a weaker relationship to earlier historical cuts. This assumption aligns with geological sedimentary patterns and significantly simplifies computational complexity, making it suitable for the real-time requirements of downhole embedded systems.

[0050] Since the actual floor height cannot be directly observed, and can only be obtained by inverting the coal mining machine's position and drum height, a noisy observation value is established: ; Among them, A k (z) represents the observed actual cutting trajectory of the k-th cut. v k ( z The noise is the observation noise, used to characterize the sensor measurement error. This noise follows a zero-mean Gaussian distribution, and its variance is... Determined based on equipment calibration results.

[0051] Step 302: The posterior distribution of the state is solved recursively using Kalman filtering. First, a prediction step is performed, and the prior statistic is output. and Then, combined with new observation A k (z) Calculate the Kalman gain K k ( z ), and then update to obtain the posterior mean. and posterior variance The posterior mean constitutes the optimal fusion estimate of the current base plate height, and the posterior variance quantifies the reliability of the estimate.

[0052] Step 303: After completing the k-th cut and updating the state, perform multi-step Bayesian prediction based on the state transition model. Due to Markov property and the closure property of linear Gaussian systems, the probability prediction of the (k+m)-th cutter's bottom plate height can be recursively obtained as follows: ; Their mean constitutes the preliminary prediction profile. ; .

[0053] To improve the robustness of the prediction model under complex geological conditions, an uncertainty threshold is introduced. and high mutation threshold If the posterior variance This indicates that the position estimate has low reliability and should not be used as a basis for extrapolation; if the difference amplitude of adjacent cuts... This indicates a geological abrupt change at that location. When any of the above conditions are met, the predicted gradient at the corresponding location z is set to zero to avoid erroneous extrapolation in unreliable areas such as faults.

[0054] Step 304, using the measured height y of the transport roadway and return airway s and y e As a boundary constraint, calculate the endpoint deviation. and Construct a linear interpolation correction function The preliminary predicted profile is then corrected to obtain the final predicted profile. P k+m ( z ); ; This correction ensures that the predicted trajectory is strictly consistent with the measured tunnel height at both ends, improving the engineering reliability of the prediction results.

[0055] Step 305: Although the aforementioned Bayesian state estimation framework can generate predicted bottom plate profiles for future multiple cuts, during the actual cutting process, due to geological non-stationarity exceeding the assumptions of the process noise model or prior information mismatch, systematic deviations may still exist between the predictions and the actual trajectories. Therefore, an outer-loop real-time update mechanism is introduced to compensate for subsequent predictions online using measured data after each cut. After each cut is completed, the actual bottom plate trajectory A for that cut is calculated. k (z) and the final predicted profile P used before cutting. k Deviation between (z) e k (z); ; This deviation reflects the systematic error of the current prediction model at position z and is stored in the historical buffer to form a deviation sequence. When generating predictions for the nth subsequent cut, an exponentially decaying weight is used to weight and fuse historical biases to obtain a compensation term. c n ( z ); ; Among them, weight λ is the attenuation coefficient; the compensation term is superimposed on the basic prediction P. n On (z), the final predicted trajectory is obtained after real-time updates.

[0056] In step 3 of this section, the process noise variance Online estimation of the sample variance of historical truncated residuals, and observation noise variance. Determined based on sensor calibration results; uncertainty threshold Take 0.04 m² as the height mutation threshold. Take 0.1 m, and the attenuation coefficient λ is taken as 0.6.

[0057] 4. Dynamic cutting base plate generation and autonomous height adjustment control module This module serves as the execution layer for navigation and cut-off planning. It integrates multi-sensor data to calculate the roller pose in real time, employs incremental modeling to generate six-DOF dynamic cut-off blocks, and forms a continuously evolving dynamic base plate model. Using a Bayesian predicted profile as the main trajectory and the height compensation amount output by the cut-off current hierarchical system as feedback correction, it generates the target roller height through adaptive weighted fusion. After kinematic constraints and temporal verification, it issues height adjustment commands, achieving a dynamic balance between macroscopic prediction and microscopic perception.

[0058] In a relatively specific embodiment, this step includes steps 401-403.

[0059] Step 401: Integrate data from inertial navigation, stroke sensors, and rocker arm tilt sensors to calculate the full pose of the coal mining machine drum in real time and determine the bottom point H at each moment. i =(x i , y i , z i The preprocessing of undercover points mainly includes two steps: First, data cleaning, which uses a sliding window midpoint filter to remove outliers caused by sensor jitter or coal block impact, while retaining effective geotechnical features; Second, spatiotemporal synchronization, which requires linear interpolation to align all data to a unified timestamp due to the different sampling frequencies of inertial navigation, travel, and tilt sensors, ensuring that the spatial position of the undercover point calculated at the same time is accurate and avoiding trajectory distortion caused by time asynchrony.

[0060] Step 402: An incremental modeling strategy for subdividing coal seams is adopted. When the coal mining machine moves to a preset spatial interval ΔS, a new geometric unit is generated. The arithmetic mean of the X coordinates of all the bottom points within this interval is taken as the center X coordinate x of the dynamic cutting block. c In the YZ plane, a Bayesian linear regression is used to fit the sequence of undercover points to the straight line y = w0 + w1z, and the coordinates of the midpoint of the line are taken as the center point (y c ,z c The slope of a straight line is converted into the pitch angle. θ p =arctan(w1); Take the average of the roll angle and yaw angle of the coal mining machine within this sampling interval, and use them as the roll angle of the dynamic cutting block. θ r With yaw angle θ y Each dynamic cut-off block consists of six degrees of freedom parameters (x). c , y c , z c , θ p ,θ r ,θ y The complete description shows that as the coal mining machine advances, it continuously generates and seamlessly splices together to form a continuous dynamic cutting bottom plate model.

[0061] Step 403: Real-time monitoring of the current signal of the drum cutting motor. Based on the physical mechanism that "cutting current is positively correlated with coal and rock hardness," a mapping relationship between current characteristics and coal seam conditions is established. The amplitude of the cutting current reflects the overall hardness of the coal and rock, the fluctuation rate of the current reflects the uniformity of the coal seam, and the abrupt change rate of the current reflects the steepness of the geological interface. The specific implementation includes the following three aspects: (1) Multi-level threshold mapping system: A mapping table between the cutting current range and the coal and rock firmness coefficient (f value) is established in advance, and the current range is subdivided into 10 levels. Among them, levels 1-4 (low current zone) correspond to pure coal cutting, and the system maintains the current height; levels 5-7 (medium current zone) correspond to coal and rock mixture or hard coal, and the system is slightly increased to avoid overload; levels 8-10 (high current zone) correspond to interbedded rock or roof and floor rock, and the system is significantly increased and the traction speed is reduced.

[0062] (2) Joint judgment of time domain characteristics: In addition to the instantaneous amplitude of the current, the time duration criterion is also introduced. If the current enters the 150-200A range and lasts for more than 3 seconds, it is judged as a stable coal-rock transition zone, and the height adjustment command is suspended to prevent malfunction; if the current is ≥200A and lasts for more than 2 seconds, it is judged as abnormal rock cutting, triggering emergency speed reduction and path replanning to protect the cutting teeth from damage.

[0063] (3) Online self-learning correction: The system records the actual cutting effect (whether it actually cuts into the rock) after each current alarm. If frequent false alarms are found under a certain current threshold, the threshold range is automatically fine-tuned. Through this "perception-execution-feedback" mechanism, the current stratification system can adapt to the differences in coal quality at different working faces and improve the accuracy of coal and rock identification. Based on the above identification results, the current amplitude and fluctuation characteristics are mapped to the local over-excavation depth estimate and the corresponding adjustment compensation amount ∆h. The adjustment compensation amount is not a fixed value, but is dynamically calculated according to the proportion of the current exceeding the benchmark value. The larger the current, the larger the compensation amount, realizing adaptive control of "yielding when encountering hard rock and chasing when encountering soft rock".

[0064] Step 404, using the final predicted profile P generated in step 3 k+1 (z) is used as the main trajectory, and the height adjustment compensation amount Δh output by the current stratification system is used as feedback correction. The target height h of the roller is generated by adaptive weighted fusion. target ( z ): ; in, a ( z ) is an adaptive weighting coefficient, which is determined by the current posterior variance of the base plate and the current value.

[0065] Step 405, perform kinematic constraint processing on the target height of the drum, and ensure that the drum lifting speed does not exceed the maximum allowable speed of the hydraulic cylinder through a limiter filter. v max ; At the same time, perform a feasibility check on the height adjustment timing. Let the horizontal distance from the current position of the coal shearer to the next cutting point be d, and the required height difference be h. According to the traction speed v t calculate the arrival time T1; ; Calculate the time T2 required to complete the height adjustment according to the characteristics of the hydraulic system; if T1 < T2, then reduce the traction speed to ; ; To ensure that the height adjustment action is fully executed, finally generate a height adjustment command and send it to the electro-hydraulic control system.

[0066] In this step 4, the adaptive weight coefficient a ( z ) has the following value-taking rules: when the current is in the normal coal cutting interval and the posterior variance of the floor , let a ( z ) = 0; when over-cutting is detected and the floor uncertainty is low, a ( z ) linearly increases to 1; when both over-cutting and high uncertainty occur, a ( z ) is limited to within 0.5.

[0067] 5. Sensor Error Simulation and System Verification Module This module provides virtual debugging and robustness verification means for method deployment. Construct an error coal shearer model, superimpose preset errors such as inertial navigation drift, random noise, and inclination sensor error on the theoretical true pose, and generate simulated sensing data to drive the digital twin scenario. By comparing the trajectory deviation and height adjustment response of the theoretical coal shearer and the error coal shearer, evaluate the algorithm's ability to suppress sensor noise, support limit testing and sensor selection reference, and reduce the risk of on-site tests.

[0068] In the relative specific embodiment, this step includes steps 501 - step 504.

[0069] Step 501, to verify the robustness of the method of the present invention to sensor measurement errors in the actual underground environment, construct an error model for generating simulated physical working face sensing data, that is, an error coal shearer model. This model generates simulated real-time sensing data by superimposing preset sensor errors on the theoretical true pose, which is used to drive the digital twin scenario and serve as the input of the navigation cutting algorithm, so as to evaluate the performance of the algorithm under non-ideal measurement conditions.

[0070] Step 502, set the sensor error model. The true pose of the coal mining machine is determined by its position coordinates (x... , y , z) Attitude angle (pitch angle) θ p Roll angle θ r Yaw angle θ y and the tilt angles of the left and right rocker arms β This section describes the common characteristics of downhole sensors. Actual downhole sensors primarily include inertial navigation systems and rocker arm tilt sensors. Their measurement errors typically consist of two components: system drift and random noise. The three-dimensional coordinates output by the inertial navigation system exhibit cumulative drift and random noise, and attitude angle measurements are also affected by sensor accuracy and environmental interference. Rocker arm tilt sensors, on the other hand, suffer from installation deviations and measurement noise. These error characteristics are modeled based on sensor calibration results and field measurement data to simulate measurement uncertainties under different operating conditions.

[0071] Step 503, let the true pose of the coal mining machine at a certain moment be... The generated simulated sensing data is the result of superimposing random errors onto the theoretical true value, that is: ; Where, Δ x Δ y Δ z Δ θ p Δ θ r Δ θ y Δ β The error value is randomly generated according to a preset error distribution pattern. This simulated data serves as input to the host computer, driving the movement of the error-prone coal mining machine model in the virtual scene, and is used for subsequent cutting trajectory prediction and height adjustment control.

[0072] Step 504 involves simultaneously presenting two coal mining machine models in the digital twin platform: a theoretical coal mining machine driven by error-free real pose, representing the ideal planning target; and an error-based coal mining machine driven by simulated sensor data with superimposed errors, simulating information from the actual physical working face. By comparing the trajectory deviation, prediction error, and consistency of height adjustment response of the two models during the cutting process, the suppression capability and robustness of the proposed method against sensor noise are intuitively evaluated. Simultaneously, extreme tests can be conducted by adjusting the error amplitude and distribution type, providing a reference for sensor selection and system calibration. This error simulation mechanism provides an effective means for virtual debugging of the algorithm before field deployment, reducing the risks and costs of actual underground testing and further enhancing the engineering practicality of the proposed method.

[0073] The coal mining machine navigation and cutting planning method based on the working face step model proposed above integrates four core modules—geological cognition, multi-step prediction, dynamic floor generation, and autonomous height adjustment control—into an organic whole. The sensor error simulation described in step 5 verifies its good robustness to actual measurement noise. In actual production of fully mechanized mining faces, it can achieve the following comprehensive application effects.

[0074] Application 1: Closed-Loop Intelligent Cutting Operation: This method upgrades the coal mining machine cutting operation from traditional open-loop control to a closed-loop intelligent mode of "perception-cognition-planning-execution-correction". The specific process is as follows: (1) The system constructs an initial working face step model based on prior geological information and establishes a three-level cognitive state division of "high confidence measured area, medium confidence projection area, and low confidence blind exploration area". (2) As the coal mining machine advances, the system automatically collects the actual cutting trajectory after each cut. This is used to construct the dynamic cutting bottom plate in real time and to update the geological knowledge status as an observation input. (3) The original medium confidence projection area is gradually transformed into a high confidence measured area with high confidence, and the newly exposed far-end area is included in the low confidence blind exploration area, realizing the gradual evolution of the geological model and enabling the coal mining machine to have the intelligent evolution capability of "cutting, learning and optimizing at the same time".

[0075] Application 2: Adaptive cutting control under complex geological conditions: This method enables the coal mining machine to autonomously adjust its cutting strategy when encountering geological abrupt changes such as faults and interbedded rock.

[0076] (1) The discretized structure of the working face step model directly represents the step change of the bottom plate, avoiding the smooth masking of the change by the continuous smooth model. (2) The current stratification sensing mechanism monitors the changes in cutting load in real time. When the drum cuts into hard rock or interbedded rock, the change in current amplitude triggers the stratification response strategy and automatically generates the adjustment compensation amount. (3) The predicted trajectory and current feedback are fused through adaptive weighting, which not only ensures macroscopic cutting along the trend of the coal seam floor, but also quickly adjusts the drum height when there are sudden changes in local lithology, so as to avoid excessive cutting of rocks or leaving coal resources.

[0077] Application 3: Digital Twin-Driven Remote Monitoring and Advanced Planning: This method maps the physical coal mining process to a virtual model in real time, providing visual support for remote decision-making.

[0078] (1) The system transmits the real-time data such as the coal mining machine's position, cutting trajectory, and current status to the digital twin platform through a communication protocol, driving the virtual coal mining machine to move synchronously.

[0079] (2) The future multi-cut prediction trajectory automatically generated based on the current geological understanding is visualized and overlaid in the virtual environment.

[0080] Operators can visually monitor the cutting status through a 3D scene in the dispatch center. When the prediction shows that there may be a sudden geological change ahead, the system will issue an early warning. When the current is abnormal and triggers safety constraints, the system will automatically decelerate or replan the path.

[0081] The scope of protection claimed by this invention is not limited to the specific embodiments described above. For those skilled in the art, this invention can have various modifications and alterations. Any modifications, improvements, and equivalent substitutions made within the concept and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A coal mining machine navigation and cutting planning method based on a working face step model, characterized in that, include: Step 1: Construct a working face stepped model and a three-level geological cognitive state model, including: The coal seam floor is discretized along the advancing direction into a sequence of transverse profiles corresponding to each cut. Each transverse profile consists of a set of discrete bottom points distributed along the cutting direction, forming a stepped model of the working face. Based on the current number of cuts completed and the preset predicted sight distance, a three-level geological cognition state model is constructed, dividing the working face into three cognition zones along the advancement direction: The high-confidence measurement area is the area that has been actually cut and for which data collection has been completed. The Zhongzhixin extrapolation area is the range that has not yet been cut off but can be predicted by extrapolation from historical profiles. The low-confidence blind probe zone is a distant region that exceeds the current system's cognitive capabilities; As the coal mining machine advances, the boundaries of the three cognitive regions move forward synchronously, realizing the dynamic migration of cognitive states; Step two, establish the dynamic evolution mechanism of a credible geological model, including: Based on multi-source static geological data, an initial discretized three-dimensional base plate height field is constructed as a priori model. After each cut is completed, the measured data is interpolated and mapped to the grid cells of the prior model. For grid cells covered by the measured data, a parameter coverage operation is performed and the confidence state is updated. Step 3 involves performing multi-step probability prediction and real-time updates based on Bayesian state estimation, including: Based on historical actual trajectories and prior geological models, the bottom plate height is modeled as a first-order Markov hidden state variable. The posterior distribution is solved recursively using Kalman filtering, and the optimal estimate and variance of the current base plate height are output. Multi-step Bayesian prediction based on state transition model is used to generate the probability distribution of future H-blade bottom plate height, and boundary correction is performed using the measured height of the roadway. An outer-loop real-time update mechanism with exponential decay weighted historical bias is introduced to compensate for the prediction online and output a future base prediction profile with probability information. Step four involves performing dynamic cutting of the base plate and autonomous height adjustment control, including: By integrating multi-sensor data to calculate the roller pose in real time, incremental modeling is used to generate a six-degree-of-freedom dynamic cutting block, forming a continuously evolving dynamic base plate model. Using the Bayesian prediction profile as the main trajectory and the height adjustment compensation amount output based on the cutting current hierarchical system as feedback correction, the target height of the roller is generated through adaptive weighted fusion, and the height adjustment command is issued after kinematic constraints and timing verification.

2. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 1, characterized in that: This includes simulating and verifying sensor errors, constructing an error-prone coal mining machine model, superimposing preset errors including inertial navigation drift, random noise, and tilt sensor errors onto the theoretical real pose, generating simulated sensor data to drive a digital twin scenario; and evaluating the ability to suppress sensor noise by comparing the trajectory deviation and height adjustment response of the theoretical coal mining machine and the error-prone coal mining machine.

3. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 1 or 2, characterized in that: In step one, the working face stepped model is defined as an ordered set: ； Where, x k Let s be the advancing position corresponding to the k-th cut. k The cross section corresponding to the k-th cut can be represented as defined in [z s ,z e Functions on ]; The working face stepped model uses actual cutting actions as an index, with no continuity constraints between adjacent sections, directly reflecting local mutations and providing a structural basis for subsequent cognitive state division.

4. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 3, characterized in that: In step two, the parameter coverage operation replaces the original prior height values ​​of the grid cells with the measured height values, and replaces the original prior variance with the sensor measurement variance. For the blind grids that are not covered, the original prior geological data is retained.

5. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 4, characterized in that: In step two, the update of confidence status involves transitioning the region that has been truncated and filled with measured data from the medium-confidence projection region to the high-confidence measured region and marking it as "credible"; the newly exposed remote region is moved from the low-confidence blind probe region to the medium-confidence projection region and marked as "to be verified".

6. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 1 or 5, characterized in that: In step three, the multi-step probability prediction based on Bayesian state estimation specifically involves: At a fixed cutting position z, the bottom plate height is modeled as a hidden state variable x that evolves with the number of cuts k. k (z), considering the random fluctuations in the bottom plate height between adjacent cuts, a first-order Markov process is used to describe its state evolution law, and the state evolution equation is established: ； Among them, w k ( z () represents process noise, used to characterize the random variation in bottom plate height between adjacent cuts caused by local nonstationarity of geological structures; Establish observation equations based on state evolution equations: ； Among them, A k (z) represents the observed actual cutting trajectory of the k-th cut. v k ( z () represents observation noise, used to characterize sensor measurement error; The posterior distribution of the state is solved recursively using Kalman filtering to obtain the posterior mean. and posterior variance ; Based on the state transition model, the probability prediction of the height of the (k+m)th blade base plate is as follows: ； Their mean constitutes the preliminary prediction profile. ; ； Introducing an uncertainty threshold and high mutation threshold If the posterior variance Or adjacent knife difference amplitude value If the predicted gradient at position z is zero, then the predicted gradient at position z will be set to zero.

7. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 6, characterized in that: In step three, the outer ring real-time update mechanism specifically includes: After each cut is completed, calculate the actual bottom plate trajectory A of that cut. k (z) and the final predicted profile P used before cutting. k Deviation between (z): The deviation sequence is stored in the historical buffer. ; When generating predictions for the nth subsequent cut, an exponentially decaying weight is used to weight and fuse historical biases to obtain a compensation term. c n ( z ); ; Among them, weight λ is the attenuation coefficient, and the compensation term is added to the basic prediction P. n On (z), the final predicted trajectory is obtained after real-time updates.

8. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 1 or 7, characterized in that: In step four, the current stratification system includes: Establish a multi-level threshold mapping table between the cutting current range and the coal and rock firmness coefficient, subdivide the current range into multiple levels, and different levels correspond to different adjustment strategies; A time-duration criterion is introduced to determine the operating condition based on the current value and its duration. The current threshold range is corrected by recording the actual cutting effect after the current alarm.

9. The coal mining machine navigation and cutting planning method based on the working face step model according to claim 8, characterized in that: In step four, adaptive weighted fusion is used to generate the target height h of the roller. target ( z ); ； Among them, P k+1 (z) is the predicted profile, and Δh is the adjustment compensation amount output by the current stratification system; a ( z ) is an adaptive weighting coefficient, which is determined by the current posterior variance of the base plate and the current value.

10. A coal mining machine navigation and cutting planning system based on a working face step model, characterized in that: Perform the method according to any one of claims 1-9.

Images